Traffic Sign Recognition using You Look Only Once Version 8 with Vision Transformer

In recent years, Traffic Sign Recognition (TSR) has become an emerging computer vision task which allows vehicles to detect and interpret road signs from captured images that facilitates for safe navigation. Even though, previous researchers have developed effective models to classify and recognize traffic signs under varying conditions, but still there are challenges such as occlusions, motion blur and luminous conditions. To overcome these challenges, this research proposes a hybrid approach using the proposed You Look Only Once Version 8 with Vision Transformer (YOLOV8-ViT) model. Initially, the data is collected from Tsinghua-Tencent 100K (TT100K) dataset which comprises of vast traffic signs in varying luminous and weather conditions. Then, the collected data is preprocessed with data augmentation and Contrast Limited Adaptive Histogram Equalization (CLAHE) for improving generalizability and visibility in poor lighting conditions respectively. Further, Convolution with Attention Network (CoAtNet) is utilized for extraction of feature which has the ability to capture spatial and long-range dependencies. Finally, the proposed YOLOV8-ViT is incorporated for TSR as it handles complex backgrounds and occlusions effectively. The proposed YOLOV8-ViT acquired better results with 98.9% accuracy, 97.9% precision, 96.6% recall, 97.5% F1-score respectively when compared with the existing TrafficSignNet model.